Proceedings Abstracts of the Twenty-Fifth International Joint Conference on Artificial Intelligence

An Approach to Cooperation in General-Sum Normal Form Games / 3984
Steven Damer

My primary research interest is social behavior for software agents to achieve cooperation in general-sum normal form games. An agent can easily be programmed to constantly cooperate in a normal form game, but such an agent is not suitable for environments with potentially hostile opponents. For a rational agent, the main reason to cooperate is to induce reciprocation; to reciprocate it is necessary to determine which moves are cooperative. In constant-sum games cooperation is impossible because any gain by one agent isa loss by the other agent. In other games (such as Prisoner's Dilemma) it is easy to identify cooperative moves because the opponent's payoffs for that move strictly dominate the other moves of the agent. In general it is not easy to identify cooperative strategies in arbitrary general-sum games.